Data-Driven Extraction of Association Rules of Dependent Abnormal Behaviour Groups Federico Antonello Energy Department, Politecnico di Milano, Via Lambruschini 4, 20156, Milan, Italy. Piero Baraldi Energy Department, Politecnico di Milano, Via Lambruschini 4, 20156, Milan, Italy. E-mail: piero.baraldi@polimi.it Ahmed Shokry Energy Department, Politecnico di Milano, Via Lambruschini 4, 20156, Milan, Italy Enrico Zio Energy Department, Politecnico di Milano, Via Lambruschini 4, 20156, Milan, Italy. MINES ParisTech, PSL Research University, CRC, Sophia Antipolis, France. Eminent Scholar, Department of Nuclear Engineering, College of Engineering, Kyung Hee University, Republic of Korea Ugo Gentile CERN, 1211 Geneva 23, Switzerland Luigi Serio CERN, 1211 Geneva 23, Switzerland This work proposes a methodology for identifying dependent abnormal behaviours through the extraction of association rules from data. The practical case considered makes use of a database of alarms generated by different supervision systems of the CERN (European Centre for Nuclear Research) technical infrastructure. The methodology is based on the representation of the alarm database with a binary matrix and the use of the Apriori algorithm for mining association rules. An application to a large-scale database of alarms generated by various monitoring systems of the point 8 of CERN is presented. Keywords: Complex Technical Infrastructures, Dependent Abnormal Behaviours, Association Rules, Alarms, Data Mining, Alarm Database Representation. 1. Introduction The methodological development presented in this work aims at improving the performance and the diagnostic capability at the Complex Technical Infrastructure (CTI) of CERN, which is the largest existing particle accelerator. This CTI is composed by several systems jointly enabling the functioning of the Large Hadron Collider (LHC) (J. Nielsen and L. Serio 2016). This is a 27 km ring of superconducting magnets and infrastructures, extending over the Swiss and French borders, and located about 100 m underground. The CTI systems and subsystems include thousands of interconnected components performing diverse functions and services, organized in complex hierarchical architectures and utilizing technologies belonging to various domains (i.e., mechanics, hydraulics, electronics, information and communication technologies). The CERN CTIs systems and sub-systems are tightly interdependent at physical, functional, spatial and, even, at data and information levels (Zio, E. 2016). For example, the electrical system of CERN CTI provides power to all the components of the CTI, whereas the cryogenic system is responsible for the refrigeration of the superconducting electromagnets along all the accelerator ring at a temperature of -271.3C, which is necessary to maintain the magnetic field required to guide and deflect the particles beams (Gentile U. and Serio L. 2018). A local malfunction or perturbation can propagate through highly dependent groups of components, originating cascades of failures across the systems possibly leading to large-scale consequences and, in the end, the CTI unavailability (L. Serio et. al. 2018). Vulnerability and resilience to failures is then an issue of concern. However, the analysis of the complex and interdependent systems, which make up the CTI, cannot be easily carried out with Proceedings of the 29th European Safety and Reliability Conference. Edited by Michael Beer and Enrico Zio Copyright c  2019 European Safety and Reliability Association. Published by Research Publishing, Singapore. ISBN: 978-981-11-2724-3; doi:10.3850/978-981-11-2724-3 0723-cd 3308